What the model does better than you — and worse

AI for storytelling and writing · Lesson 1 / 24

Division of labour, not replacement

A language model is trained to predict what usually comes next. Everything follows from that — both its power and its ceiling. It instantly produces what happens often, and almost never produces what happens once.

Where it beats you

  • Volume of options. You will not invent twenty plot turns in a minute. It will. Fifteen of them being junk is fine.
  • Structure. Breaking a draft into scenes, finding where the second third sags, showing that three chapters in a row do the same job.
  • Diagnosis. Answering "what is unclear to a reader here" more honestly than the friend who doesn't want to upset you.
  • Reframing. One idea in seven registers, from police report to ballad.
  • Stamina. It doesn't get bored on the fourth attempt.

Where it is worse than you

  • Choice. It produces options but doesn't know which one is yours. A weighted average is not a decision.
  • Specific experience. The smell of the hospital corridor in your childhood exists only in you.
  • Risk. The model gravitates towards the safe, the smooth, the expected. Literature is made of disagreement with the expected.
  • Restraint. It spells out what you deliberately left in shadow.

The substitution test

Take a paragraph and ask:
1. Could any other person with the same topic
   have written this?
2. Is there a single detail here that isn't
   already on the internet?
3. If I delete this paragraph, does anything change?
Yes / no / no means the paragraph was written
by the average, not by you.
Insight. The model doesn't understand your text, but its report on your text is still useful: it behaves like an averaged reader — and that reader is exactly who you are arguing with when you write.
Common mistake. Asking AI to write a story and judging the whole practice by the result. That's like judging a piano by the sound it makes when a cat sits on it.
Pro tip. Keep a file called 'mine' — observations, phrases, scenes from your own life. That is raw material the model does not have, and it is what makes a text unlike other texts.

Cheat sheet

  • The model is strong at volume, structure and diagnosis.
  • It is weak at choice, risk and lived detail.
  • Choice is your job, and it is what makes writing yours.
  • A paragraph anyone could have written was written by no one.
1. What is the model's main advantage over the writer?
2. Why does the model gravitate towards predictable choices?
3. What does the substitution test reveal?

🔒 Answer the question correctly to move on to the next lesson.

What the model does better than you — and worse — AI for storytelling and writing — Skilvy